I'm using the rpart package for decision tree classification. I have a data frame with around 4000 features (columns). I want to use all features in rpart() for my model. How can I do that? Basically, rpart() will ask me to use the function in this way:
dt <- rpart(class ~ feature1 + feature2 + ....)
My features are words in documents so I have more than 4k features. Each feature is represented by a word. Is there any possibility to use all features without writing them?
I figured it out:
dt <- rpart(class ~ ., data)
"." represents all features.
The caret library is really useful because you can easily apply different models and compare their performance. It can call rpart but uses a slightly different syntax to include all features.
library(caret)
library(data.table)
mt <- data.table(mtcars)
tr <- train(x=mt[,-'hp', with=FALSE], y = mt[, hp], method='rpart')
plot(tr$finalModel)
text(tr$finalModel)
Using all 4000 features for a decision tree could result in overfitting, especially if your number of observations is not huge. Caret provides built-in cross-validation. You might also want to look at model='rf' for random forests.
